Surveillance systems have applied basic object detection and facial recognition for primitive forms of functionality. However, current surveillance systems use primitive understanding of detected people. If a surveillance system is even able to detect people, it will generally only identify the location of a person as a generic person. Facial recognition can be used in limited situations but because of the recognition reliability, only a small number of people may be detected through this. Additionally, facial recognition is currently vulnerable to adversarial attacks and so is not suitable for use in many situations. This inability to understand the observed subjects provides a significant number of limitations to uses of such systems.
Additionally, large-scale applications involving computer vision are not in active use at scale. The existing solutions described above are poorly suited for such applications. This contributes in part to the lack of rich user interactions driven by computer vision.
Thus, there is a need in the computer vision field to create a new and useful system and method for linking observed human activity on video to a user account. This invention provides such a new and useful system and method.